I added an ESD next to the cantilever. I used copper wire (Lakeshore WHD-30-100) for the in-vacuum leads, and a DB9 feedthrough from KJL (FTADD09). It seems to actuate, based on being able to drive some op lev pit/yaw with the ESD channel, but I'm not confident in the op lev sensor yet due to beam divergence and clipping... not sure if we have 532 nm lenses?

I used a HeNe beam and QPD to make an optical lever with the cantilever. I'll use this to sense and damp cantilever motion, especially at its 16 Hz fundamental mode.  The HeNe beam is a couple inches lower (closer to the table) than the 1550 nm beams to allow a clear path into the chamber.

I also updated x1oma to accommodate the new channels.

I've been stalled for several days, so here's a long overdue status update.

Transmission alignment with MC2 removed

With MC2 removed to replace the cantilever, I took the chance to align transmission with 1000x more power. I adjusted the mode matching telescope on transmission path. Attachment 1 shows the transmission beam path including MC2 and the telescoping lenses (38 and 100 mm nominal focal lengths).

There is some scatter on the transmission beam path. This might be on the second transmission telescope lens, where the beam fills nearly the full lens aperture. This probably needs to be fixed, since the second telescoping lens is close to the cavity and the scatter pattern makes it hard to tell which mode is resonating.

Installed new cantilever

I used 100% IPA to remove the varnish adhesive from the flat and 0.5 m curved cantilever mirrors I've been using. After soaking in IPA for 1-2 hours, the mirror could be removed from the cantilever without applying too much force. See attachments 2-3.

I then cleaned (lens wipe with 100% IPA) both mirrors and the intact cantilever. I glued the 0.5 m curved fused silica mirror to the intact cantilever and mounted it in the clamp. See attachments 4-5.

Finally, I reinstalled the cantilever at MC2 and adjusted its yaw to see cavity flashes (I hadn't touched the input alignment, so the beam still defined the previous cavity mode).I built a more elaborate structure over the cantilever to avoid future damage. See attachments 6-8.

Overall optics setup

All of the optics we'll need for the experiment are set up and more or less aligned. Attached are some photos of the optics layout. Most of this is new since the last update, with the exception of the in-vacuum optics, pump injection path, and REFL path.

9. In-vacuum optics (cavity, Mach-Zehnder, input and output steering, transmission telescope lens). Two beams enter the chamber (pump and probe), and four beams exit (2x MZ outputs or cavity REFL beams, 1x transmission beam, 1x pickoff in the non-cavity-containing arm of the mach-zehnder). The Mach-Zehnder is intentionally misaligned after the beat note pickoff.

10. input steering and mode matching for pump and probe path. Both beams flash 00 in the cavity, but the mode matching needs adjustment. I added translating stages to the pump path lenses, but the lens alignment is imperfect; the probe path still has fixed lenses.

11. Beat note PD (1611) next to REFL camera and PD (1811).

12. Secondary MZ output PD (PDA20CS) next to TRANS camera and PD (PDA20CS).

13. The cables run along the center and one side of the table, which seems more sensible. There's not a great place to put the oscilloscope and monitors yet. I borrowed a small monitor from CTN, because one of our CCTVs died. Also, I'd like to move most of the fiber components onto the table so they can be insulated and acoustically isolated (and to avoid unnecessary loss from patch cables).

Locking attempts

Despite strong 00 flashing, I haven't been able to lock the cavity. The resonant mode is now ~18 Hz, and I think the undamped cantilever motion exceeds the dynamic range of the TeraXion laser. I used the PZT buzzer to damp the motion, but not quite enough to achieve a stable lock.

I also tried pumping out the vacuum chamber to improve acoustic isolation. However, I wasn't able to spin up our turbo, possibly due to excess condensation in the roughing pump exhaust. With the chamber valved off, the pressure is stable at several torr. The cavity misalignment due to the new index of refraction prevented serious locking attempts, and I ended up venting (twice) to realign.

Some next steps

Here are some possible next steps... The goal is to quiet the cantilever enough to lock the cavity so we can finally measure transfer functions.

• adjust alignment of pump mode matching lenses to allow z-translation and improve mode matching. Mode seems close-ish, but the cavity at 'rest' moves through so many modes its hard to tell without a single sideband sweep while locked.
• add translating mounts to the probe mode matching lenses
• adjust beat pickoff steering and show the pump-probe alignment and mode matching is close enough for us to lock the beat note
• set up the compressed nitrogen lines to float the optics table. Perhaps the cantilever below 20 Hz is driven by seismic, not acoustic
• Swap out our roughing pump and see if we can pump down (JC will join me this afternoon to do this)
• Add beam blocks around transmission telescope lens to avoid scatter into the cavity... or rebuild the telescope to avoid large beams, which unfortunately likely involves removing either MC1 or MC2

I removed MC2 and glued this 0.5 m curved mirror onto a spare cantilever. The cantilever is of low quality, but it should be fine to get the optics working.

With MC2 removed, I routed TRANS off of the breadboard and onto the table. I also routed the second MZ output onto the table along the same path as TRANS. Finally, I replaced the MZ steering mirror with a Si Laseroptik mirror with AR surface facing towards the MZ, so the beam path in each MZ arm is balanced. I routed the transmission from the Laseroptik steering mirror off of the table along the same path as REFL.

still to come:

1. Set up TRANS PD and camera on table
2. Install cantilever at MC2, recover flashes
3. Move breadboard into chamber, realign input/output optics

I aligned the cavity well enough to see flashes on a trans PD. Here's what's up next

1. route transmission beam off of the breadboard and onto the table, so I have room for a PD and camera.
2. align Mach-Zehnder
3. route REFL beams off of breadboard and onto a PD and camera
   • could lock the cavity here to make sure we're actually overcoupled, check the finesse, make sure there's no clipping, etc
4. align probe beam to pump beam. Should just involve setting two steering mirrors completely upstream of the input to mach zehnder
5. replace MC2 with cantilever, recover flashes
   • should be able to catch incident beam on the cantilever, then adjust only MC1 and MC3 while leaving input steering and Mach-Zehnder optics unchanged
6. move breadboard to vacuum chamber
7. realign input/output optics
8. lock cavity in 00, characterize

update Sat May 20 12:31:48 2023

I locked the cavity in 00 (no cantilever, easy mode, HEPA fans can remain on even). It was a little flaky and I saw some TRANS clipping (probably after the cavity).

I roughly aligned the MZ on Friday. I found some linear polarization mismatch between the two arms... maybe due to some pitch misalignment? Also, the arm path lengths aren't exactly the same, since the steering mirror on the arm-without-cavity has HR side facing into the MZ, whereas the beam entering the cavity passes through a few mm of Si before hitting HR. I'll probably replace the MZ steering mirror with a Si optic with AR surface facing into the MZ ('identical') to the cavity arm side. This also has the benefit of a giving me a <~0.1% pickoff beam in the non-cavity arm of the MZ.

I measured the angle-dependent transmissivity of three mirrors. The optical layout and measured transmissivities are attached. Since I previously measured the transmissivity of our 0.5m curved mirror at near 0 degrees to be ~168 ppm, I'll plan to use the coastline optic at 45 degrees aoi as MC3 and laseroptik_13997_2 at 40 degrees as MC1.

The notebook I used for analysis is on git (scripts/mirrorRT/transmissivity.ipynb). I did add error bars based on one standard deviation of the measured test and reference power levels, propagated to transmissivity... however these were mostly at the 1% or less level and don't appear on the plots. I doubt these capture the dominant errors, and just heuristically when I repeatedly measured the same quantity across different days, my measured transmissivities varied at ~10-20% level.

I did improved my procedure over time -- for example, by blocking the bright test beam on the test PD for the case with no DUT, I could increase the power of the reference beam and achieve a better calibration of test PD gain without saturating the test PD with the bright test beam. The coastline measurement in particular could be redone -- does 10 ppb at 45 degrees seem incredible?

It could be useful to continue characterizing more optics, but it's somewhat time consuming and I need to move on.

For ease of alignment, I'm starting the measurement at near normal incidence and moving slowly to 45 deg incidence as I realign. I focused a reference beam chopped at f_outer (f_ref) on each of the photodiodes. The test PD also has the transmitted beam.

I removed the pickoff mirror between chopper and DUT, since it didn't give me more information to have a second beam incident on the reference PD. This also seems to have eliminated most of the stray light problems (possibly I was catching AR surface reflections from this pickoff).

I took data on the coastline mirror in a jupyter notebook... haven't yet decided if it makes sense. Will upload tomorrow.

I re-ran this optimization while allowing the distance between MC3 and MC1 to vary between 1-2 inches, and found a better solution.

I repeated the measurement today with the same mirror at 45 degrees incidence and found transmissivities ~2x lower than yesterday (112 ppm). Suspicious.

I checked out the test PD signal with the mirror in path, and couldn't find the expected square wave -- only some scatter peaks at the chopping frequency.

The 45 deg mirror under test deviates the beam a couple mm, and the beam is already oversized for the test PD (PDA20CS). I added some steering and focusing lenses just before the PDs. I was able to recover clean square waves for the beam with no test mirror in path, but the beam transmitted through the test mirror always had some scatter. I added an iris between the chopper and test mirror in case I was seeing scatter from the chopping wheel, but saw no difference.

To test this, I set the mirror closer to normal incidence (~35 aoi), increasing the mirror's clear aperture. I was able to eliminate most of the clipping, but the square wave still looks rounded. The square wave shape was unaffected by the PD gain (so it's not due to BW limit on the PD).

Now, I closed the iris between chopper and test mirror, and was able to recover a clean square wave. Success! During debugging, I repeatedly misaligned the other beam paths as I moved the PD around... so I'll upload a photo after realigning the other 3 paths. They didn't present any problems after adding focusing lenses.

1. only scatter in transmission explains the nonsense transmissivity results. Paco and I did confirm a square wave yesterday, but there was still scatter. 
2. The beam was too large for the PD -- this is the case with no test mirror and no focusing lenses
3. Focusing the beam before test PD yields a clean square wave. This is the case with no test mirror.
4. Even with the beam size much smaller than the PD, the beam transmitted through the test mirror clips somewhere. Here the beam transmits through the test mirror, and I couldn't find clipping anywhere downstream of the test mirror.
5. With the test mirror closer to normal incidence, most of the clipping disappears. This suggests the beam may have clipped on the way in or out of the test mirror.
6. Closing the iris between the chopper and test mirror eliminates the remaining clipping and yields a clean square wave. The only difference between 5 and 6 is closing the iris.

[JC, Paco, Aaron]

JC brought over the cleaned breadboard. I checked that it has the right diameter. JC and I cleaned all of the optomechanics that will go in vacuum. Posts, forks, bolts, and washers were soaked in 70% IPA then wiped down. Optic mounts were wiped down with IPA and a lint free cloth. We mounted the breadboard in-air on the PSOMA table.

Later, Paco helped me place all of the optics that will go on that breadboard. We aligned the probe beam up to the output beamsplitter, and used an iris to define the beam path to MC2. We cannot align the pump path, as it is in use for the transmissivity measurements.

• Attachment 1: Before, on the "too big" breadboard
• Attachment 2: After, on the breadboard that will go into vacuum

Yesterday, I tried measuring transmissivity of a Coastline optic according to the diagram in the previous elog. However, I noticed that the PD20CS transmission PD gain knob has to be set such that the beam with no test mirror in the path does not saturate the PD, but the ~ppm transmission beam with the test mirror in path is above the noise floor of the PD. Under such a setting, square wave amplitude with no test mirror is 3.64 V, whereas with the test mirror the square wave is 6.4 mV, implying ~2250 ppm transmissivity (not accounting for power drifts of the laser). This is higher than I would expect based on the finesse of our ring cavity. However, when I used a spectrum analyzer as my pseudo-lock-in to estimate peak amplitudes, I calculated transmissivities <5 ppm (which seemed unbelievably low). I suspected that the transmission PD is not linear across these 4-6 orders of magnitude. I do observe overshooting and settling for the high-power case, and the low power case is only barely above the PD's noise floor.

To improve the measurement, I modified the setup as in attachment 1. There are now two beams chopped at different frequencies incident on each of the two PDs.

• The beam passing through the inner ring of the chopper is a square wave at f_test, and is incident on the Test PD after being transmitted through the DUT
• The beam chopped at f_test is also picked off and sent to the Reference PD. This lets me compensate for power drifts when measuring the power incident on Test PD at f_test
• A pickoff beam passes through the outer ring of the chopper, which modulates at f_ref. This beam is also sent to the Test PD, but never passes through the DUT. I've chosen the pickoff optics such that this beam's power is between the high power and low power test beams.
• The beam at f_ref is also picked off and sent to the Reference PD. This lets me compensate for power drifts when measuring the power incident on Test PD at f_ref.

The chopper generates two reference tones at f_ref and f_test (the inner and outer chopper wheel frequencies). For a given configuration (eg test mirror at 45 deg angle of incidence), I use two lock in amplifiers to demodulate the reference and test PD voltages at either f_ref and f_test. I operated the Ref PD lock in in "high reserve" mode, since by design the beam at f_test [not] transmitted through DUT is ~2-3 orders of magnitude smaller [larger] than the beam at f_ref. I used a 1s integration time on both lock-in amplifiers, which seemed to provide <1% fluctuations in the reported amplitudes while still tracking power drifts. I found that the spectrum analyzer was unable to identify the test PD signal at f_test, perhaps due to chopper frequency drift disallowing long integration times.)

The transmissivity is then calculated as

Measurements are always taken in pairs, such that P_REF at a given frequency can serve as a reference for laser power drift. DUT indicates that the beam passes through the test mirror. "0" indicates that the beam does not pass through the test mirror (DUT is on a flip mount).

Paco helped me gather the following data. We used f_test ~ 350 Hz and f_ref ~ 420 Hz (close to the maximum speed of the chopping wheel). Each time we rotated the mirror, we blocked the promptly reflected beam. We measured the case with the test mirror "out of path" three times; the first measurement (yellow) should be ignored, because we used a large integration constant for one of the lock-ins and suspect the measurement had not settled. The second and third "out of path" measurements agree to better than 1%. We were able to dump the reflected beam at 0 degrees, so that measurement should be read as "small angle of incidence" (maybe <5 deg). When we switch between measuring with/without test mirror in path, we change the gain on Test PD such that the higher-amplitude signal is ~1V. We get away with this because the Test PD sees a beam chopped at f_ref that never sees the test mirror, so the tone at f_ref always has the same power (up to drifts detected at Ref PD).

These transmissivities are believable, though I'd be happy for some feedback or skepticism about this method. I haven't carefully considered systematic errors in these measurements, and would like to quantify the error bars. Perhaps I'll gain some insight by measuring an "easier" mirror (eg 50-50 beamsplitter, so the power on Test PD doesn't change much with/without the test mirror).

[aaron, radhika, yuta]

We started setting up optics to measure transmissivity and reflectivity of the PSOMA cavity mirrors. Attached are our diagram and optics.

We first aligned the beam to irises using the x-, y-, pit and yaw adjustments on the fiber launch mount. Each time we added an optic, we adjusted the new optic's angle of incidence such that the transmitted beam does not deviate from its original path (still passes through both of the original irises). At the PBS, we will fix the last rotational degree of freedom such that the reflected beam passes through two additional irises located along the bolt lines 90 degrees from the original beam path.

The breadboard was received yesterday at noon and I began the C&B process this morning. I wiped down the breadboard with IPA wipes and cleaned the individual threads, just as Aaron and I did with the previous breadboard. Afterwards, I followed the air bake procedures from DCC-E96022-v25. there will be a 4 hour ramp up time, 24 hr bake period, 4 hour ramp down time, so we will be able to bring this to CRYO Friday morning.

Attachment #1 shows five of the IPA wipes that contained excess material left from the machining of the breadboard.

Attachment #2 shows the diameter of the new breadboard.

Attachment #3 shows the final position of the breadboard until I remove it from the air bake over on Friday morning.

We have one 0.5" fused silica mirror from Layertec, which has 0.5 m radius of curvature and nominal transmissivity ~125 ppm. I've placed this at MC2, and plan to glue it to our Si cantilever instead of the flat Si mirror currently in use. This will let us have a flat mirror at MC1, and somewhat higher finesse.

I used Shruti's HOM.ipynb to compute the cavity length that minimizes higher order mode coupling (31.4 cm), assuming MC1 and MC3 are 1.5" apart. Attachment 1 is the predicted mode scan and some relevant parameters. It doesn't look great by eye, but the closest HOM is 8th order. I also wasn't able to finish running the optimization over both cavity length and MC2 angle of incidence (laptop went to sleep and finesse crashed), so it's possible there's a slightly better solution. Will see if Shruti has tips for improving this.

I used alamode to compute the new mode matching solution for the pump path (attachment 2). I also launched the probe beam from a separate fiber, and measured its profile (attachment 2). The mode matching solution for the probe path is in attachment 3. I don't know why Matlab didn't include y-axis in the pdfs.

I steered both pump and probe onto MC2 and MC3. I'm working on the cavity alignment, with the following order of operations:

1. use input steering to align probe to MC2 and MC3. The transmits through the Mach-Zehnder input beamsplitter (BS1), so can only be aligned to the cavity axis with input steering.
2. Use BS1 to center probe on the steering mirror opposite the cavity (SM1).
3. Align pump to MC2 and MC3 using its input steering. This ensures pump and probe interfere on Mach-Zehnder input beamsplitter, since the beams are copropagating as they enter the cavity
4. Use MC3 to center cavity "REFL" on output beamsplitter of Mach Zehnder (BS2)
5. Use SM1 to overlap beams on BS2
6. Add MC1 and adjust such that beams still overlap on BS2
7. Repeat alignment procedure above to compensate for MC1 thickness, using cavity flashes to align pump and probe to cavity

approx same as before, but

• more ergonomic due to raised breadboard, core optics 4" closer to edge of table, and input beam path shifted towards edge of table
• PBS in reflection for incoming beam... the periscope mixes polarization, so I might eventually need to move the PBS after the periscope eventually.
• H45 mounts for Mach-Zehnder beamsplitters provide access to all 4 BS ports, allow 45 deg MZ angles

The purple light is from the app-enabled LED strips in the PSOMA enclosure.

To complete the Mach-Zehnder, the beam shape should evolve similarly in each arm. Since MC1 is currently curved, the arm-with-cavity has an extra F=-412 mm lens that we would need to match in the other arm. We need to place a comparable steering mirror in the arm-without-cavity, or use a flat MC1. It would also be desirable to choose different cavity optics that give a higher finesse. To that end, I'm measuring the transmissivity and curvature of several super polished mirrors on our shelves.

I'm measuring transmissivity using the Thorlabs power meter with the mirror at normal and 45 deg incidence. I'm profiling the reflected beam only for optics that I reasonably expect to be non-flat based on labeling (or lack thereof). Entries where I entered the value on the label (didn't make an independent measurement) are highlighted.

We do have one plano-concave (0.5 m roc), 12.7 mm diameter optic from layertec with T~120 ppm that could be adhered to the cantilever at MC2. We would shorten the cavity to accommodate the smaller roc (currently we use a 1m curved mirror). We would need another coating run to make a second cavity.

Alternatively, there are a number of options if we accept a 1" cavity mirror. For example, we could use the following:

• MC1 -> Laseroptik L-13997 with 330 ppm transmissivity
• MC2 -> Coastline 1m curved mirror with 200 ppm transmissivity (normal incidence)
• MC3 -> Laseroptik 4911 FS flat mirror on cantilever with 32 ppm transmissivity

The cavity would be overcoupled with finesse above 9000 assuming <100 ppm intracavity losses. This would also solve the Mach-Zehnder mode matching, since we could use a flat mirror to steer the arm-without-cavity. The FS cantilever mirror would have excess thermal noise at 123 K, but wouldn't impact the room temperature demonstration.

The full PSOMA figure of merit for the case with a 7g (1" diameter) cantilever mirror can actually be better than for a 0.5 g (1 cm diameter) cantilever mirror. The 7g mirror lowers the fundamental frequency of the inverted pendulum to ~5 Hz by increasing both mass and pendulum length, suppressing suspension thermal noise. However, the classical demonstration with only one cavity may be harder, since we rely on high gain to boost the signal above pump phase noise.

[alex, aaron]

We drew the footprint of the PSOMA breadboard (to arrive Monday) on the PSOMA table and placed the optics in-air. This will allow me to transition more easily into vacuum and handle any mode matching issues in-air.

As shown in the attachment, we placed optics for the Mach Zehnder, one cavity, and most of the input-output paths. We aligned and approximately mode-matched the incident pump path up to BS1 (input beamsplitter of the Mach-Zehnder), indicated with the red path. We included an extra folding mirror onthe IO paths to represent the 4" periscope necessary to enter the chamber. Still to come:

du

• alignment and mode matching along the paths shown in blue
• replace MC1 with the appropriate curved mirror
• replace MC2 with the cantilever
• add transmission output optics
• add probe fiber launch
• beam dumps

I'd like to buy another Newfocus 9774. These top-adjusted kinematic mounts are significantly more compact than the U100, and offer a wider clear aperture on the 4th BS port. Currently only 3/4 Mach Zehnder beamsplitters use the 9774, and adding the fourth would free up the final U100 I need for this layout.

JC brought over the He leak checker. We observed no leaks on the chamber greater than 0.4 nTorr*L/s, limited by an up-to-air valve not normally part of our system (it was on the 4-way-cross we used as an adapter to the leak checker).

[yuta, JC, aaron, shruti]

We installed two viewports on the PSOMA chamber. We first needed to thoroughly drag wipe both surfaces of each window with IPA to remove residue and dust on the new optics.

We also attempted to install the breadboard... unfortunately the breadboard was 24.17" instead of 21.47", and it doesn't fit in the chamber. Very disappointing. Expect a 1 month delay unless there's another breadboard in our labs.

I've been trying to explain the low (<50%) mode matching into PSOMA cavity. We've been observing some oddities:

• birefringence mixing ~1% of incident S-polarized light into P on reflection
• Asymmetric beam shape on reflection with yaw beam width twice as wide as pitch width
• low overall mode matching that couldn't be corrected with either alignment or lens placement

I tried several solutions that didn't work well (adding a QWP befire cavity to match apparent birefringence; shape beam asymmetrically by sending through mode matching optics at an angle to replicate the astygmatism of the input mirror; tried many different combinations and positions of mode matching lenses). I also found a mistake in my alamode simulation (didn't account for index of refraction of MC1, where we transmit through a 1m roc silicon mirror), but fixing it didn't help.

Ultimately, I think the culprit was clipping on MC1 mount. By substantially misaligning the input beam such that it is centered on MC1 and avoiding consequent clipping on REFL path optics by moving REFL cam pickoff upstream, I was able to recover a 'by eye' round REFl beam.

It's hard to realign the cavity to avoid this clipping, because MC2 is our small cantilever mirror. Since we want to move into vacuum now anyway, I'm simply going to rebuild the cavity in our chamber and be more careful to avoid clipping on this and other optics.

I had to adjust mode matching again, but quickly reacquired lock of the PSOMA cavity.

I also vented and the PSOMA vacuum chamber and began another pumpdown. Before beginning vacuum work, the chamber pressure was apparently ~utorr, but after valving off the turbo returned to ~10 mTorr. We have a second gauge on the chamber, so I'll be able to investigate the leak further...

Yesterday morning, facilities and JC came to replace the PSOMA table legs with shorter legs to enable more ergonomic working inside the PSOMA vacuum chamber.

I turned off our vacuum pump, lasers, and all equipment using the power strips under the PSOMA table. Most cables for the experiment did not need to be moved.

The operation was pretty simple

1. lift the table several inches to separate from the legs
2. Unbolt one leg at a time from the metal frame attaching to the other three legs
3. Transfer the gas piston from the tall leg (old) to the short leg (new). The TMC gimbal leveling system is identical regardless of the length of the vertical legs.
4. Install the new leg and piston in place of the old
5. Repeat for all 4 legs
6. Lower the table back onto the legs

The most 'violent' operation was lowering the table. The jacks did not allow for a gentle release of pressure, so the table had a few inches of bounce on the way down... but no apparent harm came to the cantilever or rest of the experiment.

The top of the table is now approximately level with the bottom of the horizontal bar of our 80-20 enclosure (maybe <0.5 inch of unwanted gap has developed, which could be corrected by adjusting the enclosure).

I implemented the attached loop as Rana suggested, where I used Moku to pick off and sum the boost. I haven't had a chance to finish the analog summing circuit, which will eventually save one Moku output channel.

I had an interlude where I summed the boost behind the pickoff point, creating an unstable loop... I drew some diagrams and figured out my mistake.

I also had some issues using butterworth filters... the significant phase delay within the passpand led to some passband frequencies with improved error and others with worsened error, depending on the sign of feedback. By switching to an elliptic filter with generous passband ripple, I flattened the boost's phase response and was able to suppress the error signal relative to 'no boost' throughout the passband.

I realize these dimensionless plots aren't so useful, but this is the first time the cavity lock has been stable enough to make in-loop measurements. So, more to come.

Mode matching excursion

I adjusted the mode matching for a while...

• TRANS camera has been (apparently) saturating, but I tracked down the problem to the monitor, not the camera. Plugging in TRANS cam to the other monitor displays a clean beam spot with no saturation outside of the beam. For now I'll live with a messy TRANS mon, since I was able to verify no clipping on transmission path to TRANS MON and cam.
• REFL beam is always large since MC 1 is the cavity's curved mirror. I added a FL=88 mm lens on REFL to avoid clipping the PBS. This gave me a cleaner view of the true beam shape on REFL cam.
• I had been using the second mode matching lens at almost 45 deg angle to shape the beam, but after eliminating the clipping on REFL found this wasn't necessary (beam already is round by eye).
• In the end, I have less transmission than before but also less coupling to higher order modes... so, a work in progress

Didn't take sufficient screenshots

[alex, aaron]

We connected the Wilcoxon 731A accelerometer to an SR560 running on battery power with G=100, LP at 300 Hz (12 dB/oct). We're recording the z-axis acceleration on the East side of the table for the afternoon.

Update: 4 hours after starting recording, I came back to stop diaggui and found that the SR560 battery had died (despite having started out with full charge after many days on AC line). SR560 nominally has 15 hour battery life (according to manual), so these batteries may be dying as well. Last replaced in 2017.

[alex, aaron]

I updated X1:OMA model such that boost output is only passed to the DAC if TRANS_MON exceeds some threshold (X1:OMA-ERC_THRESH). The model compiles and installs, and I've updated the medm screens.

Borrowed beam profilers and laptop from Cryo to QIL

yeah, should pick off the Moku control, go through CDS, and then sum back to the Moku control in analog. You only want to use CDS below 100 Hz, or else too much phase lag.

[dhruva, aaron]

With TeraXion laser PDH lockd to in-air PSOMA cavity, I measured the open loop transfer function and observed a UGF around 90-100 kHZ. I tried measuring the plant transfer function to ensure I've placed the PI corner at the cavity pole, but couldn't get good coherence without driving the loop out of lock. Anyway, the open loop gain looks 1/f from 1-100 kHz.

I'm setting up a parallel control path between PDH error and Teraxion control points to boost gain around the cantilever resonances when PDH is locked. Summing control signals from Moku and cds with SR560 degrades the phase margin a bit near the UGF (was only 60 degrees before), so instead I'm sending error -> cds -> moku and summing in moku.

We noticed a 2.17 MHz, ~100 mVpp tone out of DAC channels 1-4 that killed lock with boost engaged (attachment 1). This tone is not present on DAC channels 9-12 (attachment 2, actually can see some bursts of the same 2 MHz noise at lower amplitude). so for now I've just switched the boost output to those channels. This looks a bit like power supply noise (AI board is on Sorensen).

There's ~100 ms delay in the parallel boost path through cds (between picking off error and summing in boosted control). I tried a variety of filters (resonant gain near cantilever resonances, integrator from 0.01 to 100 Hz, lowpass below 100 or 300 Hz, many proportional gains, etc), but couldn't reliably reduce residual error signal with this boost. Possibly the delay is too large?

Today, I'm sending the acoustic feedback signal to a Thorlabs PE4 PZT buzzer, driving with MDT694B piezo controller. I was once again unable to reduce the noise reported by EM172 mic, but feeding back the PDH control signal led to a qualitatively quieter lock than feeding back with speakers.

Yesterday (March 23) I spent the day trying to do acoustic noise cancellation inside the PSOMA box by feeding an EM172 microphone signal back through CDS and the cryo lab speakers. Though I could see that the microphone signal was ~100x the preamp noise floor, and responsive to audio injection (eg 40 Hz tone, clapping near the mic had expected impact in the microphone channel PSD), I couldn't see substantial noise reduction (maybe 10s of % at most).

The EM172 mic isn't as sensitive as our resonant 'cantilever mic' (otherwise we'd be shining a laser at EM172), and our acoustic actuators are a couple meters and multiple layers of plastic away from the sensor. I suspect some combination of these is limiting.

Yesterday (March 22), I

• moved one microphone into the PSOMA box directly next to our cantilever. Photos are attached. There is tape rolled between the base of the post and the table, and between the microphone and the post, to provide some isolation. The cable is clamped to the table by the rubber ring around the base of the plastic box covering PSOMA cavity.
• Set up one Wilcoxon accelerometer
  • Made a 2-pin to BNC connector to read out the piezo voltage
  • Noted that if I tap the table, there is a large spike of noise (on oscilloscope) that decays in < 1 s. So the accelerometer is measuring something
  • Found one (1) SR560 that runs on battery power, and buffered the accelerometer output with a G=100 lowpass filter at 300 Hz
  • Observed the noise spectrum on an SR785 with 250 mHz linewidth. Accelerometer was oriented vertically. I didn't record the data, as I didn't integrate very long (I'll set up a couple more accelerometers in different orientations, and integrate overnight). I did note that between 1 Hz and ~20 Hz, the spectral density appears flat; there is a rise in noise between ~20 and ~100 Hz, followed by some rolloff (span was too narrow to see frequency dependence of rolloff).

[aaron, jc, alex]

We moved the PSOMA chamber body onto the optics table. The body is well centered on the baseplate, and the vacuum hardware fits inside the enclosure walls (after some minor adjustment).

Unfortunately, when we attempted to open the lid of the chamber, the o-ring had become adhered to the chamber lid. Despite the chamber being completely vented to atmosphere with valves open, we needed to use a pair of wrenches to slowly open the chamber lid. The o-ring stuck to the flat surface of the chamber lid and was lifted out of the half-dovetailed o-ring groove. We immediately saw a large gash in the o-ring, possibly from being lifted out of the sharp groove. JC had already ordered more O-rings of the correct size earlier this week.

We wiped off a large amount of vacuum grease from the o-ring, groove, and flat.

Alex took a video and photos of this process (and the entire operation from today), which are on the Cryo Lab google drive, accessible to LVK.

[jc, aaron]

We received a new o-ring for mating the baseplate and body of the PSOMA chamber that we think is now the correct size: 22.94" ID x .275" W (ACT) or 23" ID x 1/4" W (NOM). Part number from KJL is O-V472.

We removed the too-large o-ring, wiped down the groove and new O-V472 o-ring, and installed the new o-ring into the groove. Looks like a good fit! JC is posting photos shortly. We also took a video scan of the o-ring to show that there are no cuts, gashes, dents, etc.

Chris implemented a python-based (pcaspy) EPICS IOC and driver to communicate with the TeraXion laser. After running teraxion.py from the windows laptop's command line prompt, all associated EPICS channels can be modified from the Linux machines on cryo lab network. In particular, X1:OMA-TERAXION_OPTREF_TEMP can be modified to change the optical reference temperature, which provides slow, wide range detuning of the laser frequency.

To startup the server, I ssh'ed to the laptop (15754@10.0.5.240) in a tmux session on spirou, then ran 'python teraxion.py'. I made a new .ini file for slow temperature control of the TeraXion pump, using X1:OMA-TERAXION_OPTREF_TEMP as the actuator channel.

After much mode matching effort, I was able to lock TeraXion to the cavity for ~minutes with slow temperature control engaged. However, when I tried to adjust mode matching I lost lock and was unable to reacquire.

I found the EM172 microphones at the 40m, but couldn't find the appropriate circuit board. I found something close to the photo in 40m log 6651, but it's not identical and I found later mention of a broken EM172 microphone removed from PSL.

Rather than test the mystery board, I filtered PDH control with a Q=3 resonant gain filter at the cantilever fundamental frequency and fed that to the speakers. After adjusting the digital and speaker (analog) gains, I was able to reduce residual noise at the cantilever frequency by a factor of 2-3. This was sufficient to maintain cavity lock for more than 10 minutes.

I also locked the N (probe) laser to TeraXion. I injected a signal at 99-160 Hz and identified the same peak on PDH error and TRANS MON channels, but wasn't able to maintain lock with high signal amplitude long enough to measure transfer coefficients.

On Monday, I removed the label tape damping PSOMA's cantilever (MC2). I also installed a ~56 Ohm Omega heater onto MC3's post, and wrapped the post in 3 layers of foil. I applied \sqrt(2.5^2 / 2)~1.76 V across the heater, to provide both positive and negative actuation range for the cavity's slow thermal control.

I managed to lock TeraXion to the cavity, but tracking the undamped cantilever's fundamental (~40 Hz) mode requires almost the entire Teraxion actuation range.

Today, I tried sensing acoustic noise with an optical lever on the plastic box covering the cavity, filtering the noise around the cantilever mode, and feeding back with the speakers to damp the cantilever. Didn't work very well.

I instead improved the acoustic isolation provided by the box by clamping more of its sides and adjusting it such that rubber sits between the edge of the box and the optics table on all sides (at the expense of clipping TRANS beam on the box itself). With these modifications, I was able to lock the cavity with no acoustic feedback for ~10 s.

The optical posts we purchased from Thorlabs are RS3.5P. I have drilled holes into the sides of these to allow ventilation. (Showed in Attachment #1) I used a 1/8 drill bit for this. The posts are currently Sonicating in the C&B lab for 10 minutes. (Shownin Attachment #2). 

[JC, aaron, paco, anchal]

JC and I measured where the PSOMA chamber will sit on the optics table. We confirmed that the lid will not impinge the planned shelf at the center of the table, both CF flanges have sufficient clearance from the edge of the table (4 rows of bolt holes), and the vacuum valves at the NW flange have sufficient clearance from the walls of the enclosure.

Paco joined us to lift the chamber body with the hoist, and acted as a third hand on the hoist as JC and I moved the baseplate into position on PSOMA table. We covered the open bottom of the chamber with UHV foil.

JC and I proceeded to remove the old o-ring that we suspect is damaged. We cleaned the o-ring groove with IPA and lint free cloth. While cleaning the groove, we found metal filings (mm size) on the inner bottom edge of the groove, which we removed with a magnet wrapped in lint-free cloth.

We wiped down the new o-ring. Before placing it in the groove, we suspected it is slightly oversized, and indeed we were unable to install the o-ring without a bubble of extra viton. The new o-ring is nominally undersized (608.08 mm diameter o-ring for a 613.1 mm diameter groove)... we will try again this afternoon to seat the new o-ring, using IPA as lubricant to let the o-ring relax into the groove as we go.

Paco and JC had to leave, so Anchal arrived to help me lower the chamber body onto wooden blocks.

I turned on the N laser before leaving so it would be warm today. I'm seeing 20 dB less RF power in the pump-probe beat than I had been when using the Rio S laser as pump, consistent with using a 99-1 BS while keeping total optical power constant (TeraXion power incident on BEAT PD is about the same as previous pump, due to new BS; probe power is 1/10 what it was before).

I locked both PDH and PLL. I'm keeping TeraXion in 'autonomous' mode, and adjusting its optical reference temperature manually to maintain lock. The cavity drifts through the full range of TeraXion's analog frequency tuning in <10 min, so there isn't a lot of time to adjust pump-probe phase and observe signal transfer functions. Is there a Windows™ way to remotely/automatically interact with its GUIs? Short of adding some slow control to one of the cavity mirrors, I'm not sure how to keep TeraXion locked for long durations.

After about 7 hours of locking effort, I did get 60 seconds of clean signal injection. Not a good ratio. Could probably extend this by standing next to the laptop and manually holding PDH control signal near 0.

issues, comments

•  lines in the beat note spectrum at ~135 kHz (possibly just loop oscillating) and ~380 kHz (seems to be associated with TeraXion). Attachment 1.
• 40 Hz cantilever mode amplitude oscillates at ~1 Hz. During periods of high cantilever mode amplitude, PDH loop 'flickers' at 40 Hz extrema. I tried to identify the oscillation frequency, best guess is ~500 Hz (near cantilever's second harmonic).
• Pump and probe frequency response are now very dissimilar, complicating the PDH-to-probe feedforward scheme. However, since the PDH loop is much more stable with the extra power, I was able to disable the feedforward and still maintain PLL lock. Problem averted.
• I fed the PDH error signal to the speaker to damp the cantilever mode by a factor of 2-4 (amplitude). Mostly just for a break. Previously I'd done this with moderinger (x1siq) model, which of course works better due to the filtering and etc

We need about 10x more laser power to see amplified signal above the pump frequency noise floor. I'm replacing the "S" Rio laser with the TeraXion laser to get this factor.

After the Windows laptop completed an automatic security update (should disable, but I don't have sufficient credentials), I was able to launch PureSpectrum and connect to the laser driver.

differences in using TeraXion laser

• I don't think there's a way to remotely tune laser temperature through PureSpectrum
• No analog current (or frequency) monitor. The LIGO current drivers have a DB9 output providing HF, LF, and Cathode monitors. Instead, I'll pick off Moku's PDH control output.

optical path power checks

The TeraXion laser is ~80 mW from the diode. Here are the maximum optical power ratings and losses for the components along its path. I've highlighted where I need to make changes.

Looks like I'll need to send even less optical power to the BEAT PD. I have an unused 99-1 fiber BS, so I can replace the 90-10 pickoff BS with 99-1 and stay well below PD saturation.

Even after an extra patch cable and some 0.5% reflectivity at the fiber-to-free-space coupler, I expect ~14 mW from the pump in free space. All of the free space optics can handle 14 mW, and to avoid saturating REFL PD (1811) I used the HWP/PBS to direct most reflected power to the camera, and placed an OD 1.0 reflective ND filter in front of the camera. I dumped the ND filter reflection on a new razor dump.

I only observed ~5 mW launched to free space when I first turned on TeraXion (after making the appropriate fiber connections). I'm missing a factor of ~3 in pump power incident on the cavity, and a factor of ~2 20 incident on BEAT PD. I measured pump power at a few fiber connection points filled into the table above, and after cleaning some fibers recovered >8 mW launched. I can do a more systematic fiber cleaning with the laser off after locking.

ways to get more pump power

• move fiber EOM and 50-50 BS to optics table, eliminates 3 patch cables on the pump path (1 dB excess loss)
• Mix pump and probe in free space. This is the plan when we implement Mach Zehnder. Saves ~10% excess loss
• Free space EOM, at least for pump
• Thorough fiber tip cleaning

Together these could recover a factor of 2-3 in pump power.

Current feedback to TeraXion, high power cavity locked

Following the TeraXion user manual, I limited Moku's PDH control output to 2 Vpp. The manual specified that TeraXion can handle +- 2.5 V in 'manual' mode; I'm actually operating in "automatic" mode, and it says to consult the test data that came with the laser (which I don't have, or at least don't remember having seen). In "manual" mode, modulation signals should come from a floating output source to avoid 60 Hz noise; it doesn't specify whether this is corrected in "automatic" mode. Lastly, the input impedance of TeraXion's mod input port is only 10 kOhms, so they recommended driving with a <50 Ohm source.

I connected Moku directly to Teraxion modulation port with a 20' SMA cable. After some effort, I acquired lock. The PID filter looks about the same as before, but with some extra roll up at low frequency.

TRANS CCD camera is saturating, I should add a filter there too.

tldr: autolockers can now reacquire PDH and PLL lock, so I should be able to record data for O(days) and integrate the signal transfer function offline after postselecting for periods of stable lock.

Chris pointed me to 'cdsutils switch' and ezca. Thanks!

I modified autolockPSOMA.py to switch LIGOFilter banks with ezca.

I also slightly changed the logic of the script. 40m's MC autolocker scanned by activating a lockin block in rtcds. Since I just want to handle slow temperature control with a python-based PID, I'm incrementing TEMP_TUNE directly in autolockPSOMA.thresholdSearch.

I also introduced a set of scan states, which are set (unset) when a scan starts (stops). This is to avoid the slow temperature control PID from interferring with the autolock ramp. Autolocker disables slow temperature control when scanning starts, enables it when scanning stops, and sets all slow temperature control parameters once lock is acquired. The slow temperature PID script still needs to be running in a separate process, autolocker only handles the EPICS channels.

The autolocker for PLL requires that the PDH loop first be locked.

I tested that when both PDH and PLL lose lock the autolockers can successfully reacquire lock. The autolockers approximately follow the steps outlined in my elog above, but don't wait for the slow pump-probe phase loop to settle before turning on signal.

We'd like to integrate PSOMA's signal transfer functions for longer, both by extending periods of continuous lock and quickly reacquiring lock then postselecting on periods of 'good' data.

Anchal has pointed me to the autolock scripts that (I think) Yuta developed for the 40m MC cavity (40m/scripts/MC/autolockMC.py, */ALConfigMC.yml). I've copied the scripts into cryo_lab/scripts/autolock so I can update them for our experiment.

I should be able to automate all of the steps I take to reacquire lock when both PDH and PLL lose lock:

1. Turn off f_PLL modulation that generates on-resonance probe sideband. Until the PLL acquires lock, this sideband will only break PDH lock.
2. Turn off signal excitation
3. Turn off pump-probe phase servo, and reset its integrator
4. Step probe TEC to detune probe from pump by >~GHz
5. Sweep pump TEC in both directions until PDH locks
6. Sweep probe TEC (opposite direction from step) until PLL locks. If probe frequency 'overshoots', reverse sweep.
7. Turn on f_PLL modulation to generate on-resonance probe sideband
8. Adjust PLL LO phase in Moku to minimize error on pump-probe phase loop
   • This requires GUI or digital interface with Moku. I'm trying to avoid remote comms with Moku, since I haven't set up the infrastructure.
   • However, Marconi's PM sensitivity is up to 10 rad/V, so I should be able to make this adjustment using only Marconi's DC-coupled mod in port without using up too much dynamic range.
   • Technically this could be done by simply turning on the pump-probe phase loop, but it might be more efficient to sweep?
9. Engage pump-probe phase loop and wait for it to settle
10. Turn on signal excitation

Here's my understanding of the structure of the script, which I've color coded to correspond to my steps above. Generically, these consist of preparing the system to acquire lock, scanning to acquire lock, and preparing the system for measurements.

• Blinks while the autolock script is running
  • While the autolock is enabled and the loop is locked (unlocked) based on monitor/threshold when you want (don't want) lock based on required states
    • If you want the loop locked, do nothing. Loop is already locked!
    • If you want the loop unlocked, prep for eventual locking by disabling all of the loopState and loopStateDisableAux channels (plus some hard-coded extras for some reason)
  • If you're instead in the 'wrong' state while autolock is enabled, the autolock script needs to kick in
    • If you want the loop locked based on required states
      1. Prep for locking by disabling loopState and loopStateDisableAux channels
      2. Scan until the loop is locked
      3. Update loopState cahnnels to 'enabled' state (plus some hard-coded settings)

If I avoid the hard coding, I should be able to use the same script to lock both PDH and PLL but use two different config files.

The flags (monitor/threshold pairs) I'm using to indicate lock:

• PLL locked when PLL_Q (directly from Moku) > some threshold, currently 500 counts
• PDH locked when TRANS_MON > some threshold, currently 500 counts
• pump-probe phase locked when lockin I error is < threshold (something like < 1 deg of phase error, where the conversion to 'counts' changes depending on where I stick the loop gain). This loop doesn't need an autolocker, it might be sufficient to turn it on as long during analysis I check whether the transient settled. It's "locked" whenever the PLL is locked.

I'm having trouble switching toggles within filter modules like X1:OMA-[SYSTEM_FILTERBANK]_SW2 on/off. Writing 1 or 0 to these channels instead toggles some combination of the switches available on a particular filter module. I've worked around this before by hard coding the appropriate value to write to *_SW*; for example, to turn the output of a filter bank on/off, write 1024 to *_SW2; the corresopnding readback channel *_SW2R changes between eg 512 and 1536. However, since this always changes the state of the switch, I would need to check the state each time. I'd like to avoid this check, since I don't know for sure that all filter banks report their state consistently.

I noticed at 40m the corresponding switches in autolockerMC have Enable/Disable states, rather than separate write/read channels as in cryo lab. I'll ask around for how these are set up at 40m.

I've been noticing that when pump and probe are detuned by 300 MHz, I see beating on REFL MON even when the Marconi output is off. Something must be mixing the 300 MHz pump-probe beat note back down to DC on REFL.

I started disconnecting cables to eliminate potential sources of cross talk.

• Turned off PDH and PLL servos before control filter in Moku interface
• Disconnected Marconi from AM mixer, no change
• Disconnected AM mixer RF output from N EOM, no change.
• Disconnected Moku from HF mod ports of laser current drivers, no change.
• Disconnected OCXO from S EOM, no change. This is the last fast phase modulator, so I'm pretty confident there isn't a 'real' pump or probe sideband at 300 MHz that I don't know about causing the beating.
• Changed LO frequency for internally generated Moku oscillator by 100 MHz. Beating goes away, returns when pump probe are detuned by only 200 MHz. Good indicator that Moku is sending 300 MHz somewhere it shouldn't be going.
• Disconnected PLL 'lock indicator' from the BNC-to-DB9 board also carrying REFL_MON, TRANS_MON, and BEAT_MON. beating stops. The beating also stops if I disconnect the DB9 cable from BNC-to-DB9 board.
  • The PLL 'lock indicator' is the Q output from the PLL mixer. Apparently the PLL's lowpass filter passes some 300 MHz when the loop is unlocked, which then leaks into REFL_MON in the DB9 cable (which is wrapped around a ferrite toroid).

I'm also concerned about connecting REFL_MON to an oscilloscope and a long DB9 cable without buffer to protect the PD.

I rerouted TRANS_MON, REFL_MON, and BEAT_MON first through SR560, <30 kHz, 24 dB/oct buffers before any other scopes or instruments. Doing so eliminated the 300 MHz coupling to the oscilloscope displaying REFL_MON, and also eliminated a ~1 Hz oscillation I'd been seeing on REFL_MON and TRANS_MON (probably a ground loop).

adjusted control filters

I adjusted the PDH and PLL control filters to increase time between lock loss.

The PLL in particular had gain peaking around 20 kHz and 200 kHz, and I eliminated both by reducing the integrator corner (lowered frequency at which phase margin becomes 0, corresponding to higher gain margin at this frequency) and reducing proportional gain (increasing phase margin at the UGF). See attachment 1.

For the PDH filter I increased proportional gain while decreasing the integrator corner. See attachment 2.

signal transfer function (again)

I identified that TRANS_MON is relatively quiet around 113 Hz (quieter regions at higher frequency have noise scaling worse than 1/f^2 from 113 Hz). I'm injecting signal at 113 Hz appears about 4-5x higher than the noise floor in TRANS_MON.

I'm running into some errors in nds... Diaggui is failing with error "test timed out". Ndscope reports all-NaN slices. I restarted rtcds, which cleared a TIM error on X1IOP_GDS_TP, but didn't resolve the nds errors. It looks like the models are still running, data just aren't available to nds.

I used systemctl to restart rts-nds.service, rts-edc, and rts-daqd, and the problem is resolved.

Found another bad BNC cable that was preventing sensible slow control of pump-probe phase.

Eventually, found the correct sign and gain of the slow pump-probe phase loop to keep the pump and probe aligned (90 deg out of phase).

Still having trouble driving the cavity without causing drift in the slow pump-probe phase.

I once again measured a sign response with settings as in attachment 1. The cavity and beat note remained locked throughout O(20 min) measurements. Spent most of the day measuring transfer functions, never got good coherence in the HF_MON channels.

I continued shelving SiFi optics to make room on PSOMA table for the new vacuum chamber. Also showed Yuta around so he can start a silicon birefringence measurement. 

I was right earlier in the week, DC errors between Marconi and DAC ground translate to frequency offests in external FM mode that are not corrected by the slow pump-probe phase lock loop.

I switched the Marconi to PM mode and added a digital integrator in cds. I manually nulled the 10 kHz tone in BEAT_MON by adjusting PLL LO phase. Then, I increased the Marconi PM sensitivity to maximum (10 rad/V) and was able to correct for small (deg) phase errors introduced at PLL LO.

With ~2 minute integration time, I measured a sine response at 45.375 Hz (close to one of the cantilever resonances) with close to 1 coherence in all monitor channels. Averaging over multiple observations decreases the coherence, though I'm not sure why. I'll try for a swept sine next week, keeping in mind this will get immediately easier in vacuum with a lower acoustic background.

I measured the transfer functions relevant to the PSOMA signal transfer functions. Results are in /cryo_lab/data/TransferFunctions/20230210/signal_TFs.xml. From high to low frequency, I lost lock twice; at the discontinuous steps around 400 and 600 Hz, I realized the loop controlling slow pump-probe phase was oscillating, so I reduced the gain.

The transfer functions to BEAT_MON, TRANS_MON, and REFL_MON have mostly high coherence throughout because these are sensitive to amplitude fluctuations around the pump (the signal is injected into amplitude quadrature). NLD_HF_MON and SLD_HF_MON (the control signals driving NLD and SLD currents respectively) have low coherence where PSOMA gain is low, since the optomechanical transfer function is responsible for rotating and amplifying AM into PM.

I suspect we see low coherence to both AM and PM monitors near the first cantilever resonance (~43 Hz) because here PDH loop gain is sufficiently high that signal appearing in SLD_HF_MON moves laser frequency away from resonance the noise floor is too high due to residual cantilever motion. I suspect there is another cantilever mode just above 100 Hz that might be in a 'sweet spot' for signal amplification without moving the cavity away from its set point, since coherence to NLD_HF_MON and SLD_HF_MON increases without losing coherence to AM mons. [update: tried measuring again just above 100 Hz, couldn't repeat apparent coherence in HF mons]

To measure these TF, filtered the excitation with a derivative between 10 Hz and 1 kHz. To normalize these transfer functions in terms of relative AM, it makes more sense to put TRANS_MON(f) / TRANS_MON(0 Hz) in the denominator (transmission AM relative to mean cavity transmission).

I'm updating the PSOMA cds model to handle slow control of pump-probe phase. I'm using a lock-in I found in rtcds/userapps/release/cds/common/models/lockin.mdl to drive and demodulate probe AM. I sum the lock-in's in-phase signal to the excitation channel that FMs the Marconi RF.

I also realized that when we eventually want to do feedforward nulling of signal excitations apperaing in PDH (or PLL) control signals, we can use the low-frequency 'from DAC' port of the current drivers. I've added some channels to handle this, but won't implement anytime soon.

Lastly, I cleaned up some unnecessary channels related to slow TEC control (which we want to handle with a python script).

The top level view of the model and BEAT subystem are attached.

After saving the model, I make'd, installed, and rtreset the models on cymac1. Then, I ran mdl2adl.sh on cominaux to autogenerate medm screens. Note that I needed to actually run the cronjob script, since otherwise the old medm files aren't cleared out.

The cables situation was starting to bother even me, so I removed ~20 unused cables from the system and labelled most of the remaining if they have permanent positions.

After the seminar, I confirmed I'm still able to lock ERC+BEAT (east ring cavity + pump-probe beat phase). Temperature control seems to work well without acromag glitches, though i'm still seeing some unexpected lock loss after ~10s min.

I measured the attached power spectra and transfer functions. These are uncalibrated and mainly to demonstrate the salien features:

• 10 kHz tone is absent from BEAT MON, because I'm controlling pump-probe phase such that AM entering the probe EOM appears as pure PM around the pump
• I am set up to measure transfer functions, but still have very low coherence (same place I was at on Tuesday)
• Acoustic modes everywhere, cavity is doing some filtering, HF mon for S laser reflect PDH control signal while HF mon for N laser reflects PLL+PDH control signal, etc

I can try integrating longer to boost coherence in these transfer functions, but ultimately I think I need to inject a larger signal. I was worried that by using the Marconi in externally driven FM mode, any output offset from the DAC is translated into a frequency error on the Marconi. If we control the lock-in output to be 0 counts, and if 0 counts on the lock-in is not equivalent to 0 V into the Marconi, we introduce a DC pump-probe frequency error! Fortunately, the 'error signal' in this loop is defined at the Marconi input, not at the lock-in output... a difference between voltage into the Marconi and Marconi ground cause the pump-probe phase to slip. Phew, safe to increase Marconi FM sensitivity (Hz/V).

I tried to increase Marconi FM sensitivity, but couldn't lock the phase-adjusting loop. Until tomorrow.

I made a ~ 1.5 Hz low pass filter with a 11 kOhm resistor and 10 microFarad capacitor in a Pomona box to place between the DAC and the TEC, used for slow temperature tuning, to avoid the weird voltage railing that happens when the PID output changes by small values (especially around 0 V) from reaching the TEC.

Attached is a quick transfer function I measured with teh Moku:Lab.